Imported from paniolo-ai/scan (
AGENTS.md). Install upstream withnpx skills add paniolo-ai/scan. Copyright stays with the author.
AGENTS.md
Repo-wide instructions for AI coding agents working in paniolo-ai/scan — the public,
open-source adapter layer that lets a coding agent run the deterministic paniolo scan CLI
(via @paniolo/cli) and remediate its findings.
Read First
- [[sharp-shooter-wiki:rules]] — canonical project rules (this file defers to it).
- [[paniolo-wiki:available-skills]] — skill slug index.
- README.md — user-facing install and harness setup.
Thin adapters such as CLAUDE.md and .agents/README.md should stay short and point back here.
What this repo is
This repo ships only thin triggers around paniolo scan (via @paniolo/cli). Rule
definitions, severity thresholds, scores, and grades live in the compiled paniolo scan CLI,
not here. It exists
to expose two public distribution channels: the portable skill (installed via npx skills add,
the skills.sh listing) and the GitHub Action (a CI gate).
The three flow surfaces all describe the same interactive scan → present → remediate → re-scan loop:
| Surface | Path | Consumed by |
|---|---|---|
| Portable skill | .agents/skills/paniolo-scan/SKILL.md | Any skill-capable agent (via npx skills add) |
| Slash command | .agents/commands/paniolo-scan.md | Claude Code plugin |
| Workflow | .agents/workflows/paniolo-scan.md | Antigravity / Gemini |
The GitHub Action is a separate, non-interactive surface — it runs the diagnostic only (no remediation) as a CI gate that fails the build on findings at or above a threshold:
| Surface | Path | Consumed by |
|---|---|---|
| Composite action | action.yml | GitHub Actions CI (uses: paniolo-ai/scan@<tag>) |
| Action self-test | .github/workflows/action-self-test.yml | Proves action.yml runs end-to-end on a real runner |
Core Rules
Treat [[sharp-shooter-wiki:rules]] as canonical. If this file and the rules doc disagree, the rules doc wins.
- Never reimplement scanner logic here. Read rules, scores, and thresholds from the CLI's JSON report; do not hard-code or duplicate them in an adapter.
- Keep the three flow surfaces in sync. A change to the scan/remediate flow in one (SKILL.md, command, workflow) should land in the others, adjusted for that harness's tools.
- Keep the GitHub Action in sync with the CLI contract. When the CLI's flags or severity vocabulary change, update action.yml inputs, the README's Action section, and the self-test workflow together.
- Keep
SKILL.mdportable. It installs into other people's repos — it must stay self-contained, with no links to paths that exist only in this repo. - Always preserve the goodwill framing. Every flow surfaces it once before remediating, and it never gates the free remediation.
- Keep adapters thin. Put reusable guidance in this file or
.agents/, not duplicated across adapters.
Validation
After editing markdown, format and lint it (if tooling is configured), then dogfood the scanner on this repo:
npx @paniolo/cli scan --format json .
The repo should keep a clean meta-harness profile — it is the public face of a product that measures harness quality.
Safety
The scanner is diagnostic and read-only. All file writes happen in the user's working tree and are the agent's responsibility, never the scanner's.